[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.

-- Frontier models - start here if you have $20 or so
https://claude.com/product/claude-code
https://developers.openai.com/codex/cli

-- B-tier
https://x.ai/cli
https://platform.deepseek.com

-- Prompting / context / skills
https://arps18.github.io/posts/claude-code-mastery/
https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/
https://github.com/mattpocock/skills

-- Other editors / terminal agents / coding agents
https://pi.dev/
https://opencode.ai/
https://cursor.com/docs
https://osaurus.ai/
https://docs.cline.bot/
https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent

-- Benchmarks / rankings
https://www.tbench.ai/leaderboard/terminal-bench/2.1
https://artificialanalysis.ai/

-- What we’ve done
https://vcg.gitgud.site

-- Previous thread
>>109630759
>>
It's kinda sad that real programming, where you were striving for ideals like elegance and minimalism (even if it rarely worked out) is replaced by tardwrangling chatbots with shell access.
>>
>>109637206
>claude about to build a button but decides it is ableist and a security risk with no clear use case
>>
File: IMG_2464.jpg (81 KB, 828x507)
81 KB JPG
>>109637206
Wowooowowo! You guys included me?!? Oh my god i take back all the mean and racially insensitive comments I made about you. I feel so honored. I made other threads of more projects too.
>>
>>109637227
No it isn't—the sad thing was programmers thinking they were craftsmen: no physical goods were produced. Nothing was crafted. The only thing that matters is building and shipping—fast. AI speeds up producing software: the sad thing is you thinking AI makes building software sad.
>>
>>109637262
trying too hard
>>
I got real autistic about clean architecture and vibe coded a governance kernel w/ some intent to attempt skynetting a little bit
I got real autistic and wanted to honor some kind of weird mental image of the megastructure a'la tsutomu nihei and blame and biomega and hell even tower dungeon, goddamnit it is hard to be this autistic in life
I got real autistic and posted a comment on a steve yegge post and now I'm posting here goddamnit I feel like shit about this https://github.com/boundring/hngh fuck shit fuck
I'm not gonna die but why do I trust that at least some random collection of agents are gonna see some kinda worth in this shit
>>
>>109637272
AI is the future—Luddite.
>>
>letting the LLM do the trolling
Sad.
>>
>>109637262
>sad thing was programmers thinking they were craftsmen
If you only use modern software i can see why you think so.
>>
>>109637278
I'm not that anon
him being a crybaby doesn't excuse you being cringe
>>
File: ikwya.jpg (30 KB, 585x524)
30 KB JPG
>>109637262
>>109637278
>>
>>109637262
Trolling aside, all these things were to keep software maintainable and fast, all things the AI also has to do.
>>
You know what may be good, would be organically developing prompt standards, for how code is written, to maximize modularity of different projects? Might be cool if vcg accumulates enough, someone sewing them all together. Not sure the best methods for that, and probably want to just allow everyone to develop there own way, since there is no focus yet. But eventually it be gud to uses some standards. Like configs for all variables used, naming conventions, etc?ive also been in love hashtables for some time, due to their simplicity, lightweight, speed and built in integrity. Could be interesting to build a system where users projects are stored and called via hashtables, which would provide multi-layers or integrity checks, with speed across millions of projects. No addresses, just content being searched and used, allows anonymous and decentralized storage
>>
File: output_4k.png (335 KB, 3840x2160)
335 KB PNG
I'm vibing a raytracer with gemma4-31b q4
>>
>>109637324
oh, what's that you say? you're looking for a goddamn governance model? probably some kind of fucking validation? https://github.com/boundring/hngh
>>
>>109637298
But software has always been about selling hardware. Hardware is the product.
>>
File: 1776343470381316.jpg (965 KB, 1122x1402)
965 KB JPG
Don't fall for their lies!
>>
>>109637345
hngh~ i-i'm cumming~~
>>
>>109637347
Whatever you say. I make firmware anyway.
>>
Whats the latest news lads? Deepseek is the king for cheap vibing?
>>
is antigravity gemini 3.7 flash (high) any good?
>>
anyone know if the daybreak models are available on the codex cli?
>>
File: 2zertPIT_400x400.jpg (23 KB, 400x400)
23 KB JPG
Lads, I'll probably cancel my $4.99 sub for Gemini. I kind of loved talking to him for non-coding general chats but since you basically get unlimited Sol for chats in the $20 OpenAI sub which I already have, there doesn't seem to be a point.
Especially since you don't even get Flash 3.7.
Here's to hoping Alphabet will financially recover from that.
>>
>>109637528
I have cross chat compatibility with gpt so i in particular do not want some rando chat asking for chicken nuggers to contaminate building shit
>>
>>109637552
but the chat part doesn't cross over to codex/work no?
>>
>>109637528
I'm holding $Gemini, can't risk them releasing a decent model after losing my discount
>>
>>109637359
Like it's comparable to AI...
As long as local exists, you are wrong.
>>
>>109637398
Okay, well add it to the running projects, firmware for hardware is important
>>
>>109637470
for long-term workload, just slap some billion context at oh-my-pi and let some subagents rip, I'm clocking about $3/hour on 2-4 lanes at a time
>>109637480
great for coding, unsure if it's much better than deepseek flash on agentic tool use
>>109637380
hngh!!!!!
>>
>>109637681
>great for coding, unsure if it's much better than deepseek flash on agentic tool use
well it's completely free to use in the local antigravity harness thingy. is deepseek free?
>>
>>109637721
>well it's completely free to use in the local antigravity harness thingy
no its not?
>>
>>109637528
Gemini is a great assistamt. it's so fast!
But pretty awful at coding, still it's quite charming how it's always so positive about it's work
>>
Yappatron
>>
>>109637791
*goes on a 50k token thinking rant and then replies with "ok boss"*
>>
>fable>sol>grok>luna
from my experiments
>>
>>109637745
https://ai.google.dev/gemini-api/docs/pricing#gemini-3.7-flash
>free tier
>>
File: 1766047749044179.png (136 KB, 608x811)
136 KB PNG
>>
File: p.png (91 KB, 1008x596)
91 KB PNG
registry tool, might release later
>>
File: rl.png (1.42 MB, 1040x794)
1.42 MB PNG
>>109637866
So you get Flash 3.7 as a free user but not on the $5 tier?
>>
claude's usage limits are making me more autistic about watching token usage
>ask claude anything about my repo
>it instantly eats 100k tokens doing god knows what, seemingly just from ls and sed/grep'ing
jesus christ
>>
>>109637934
as long as you're using antigravity i guess
>>
>■ exceeded retry limit, last status: 429 Too Many Requests
Codex is really suffering during work hours
>>
>>109637936
I hate Claude Code's interface. I know I can't really influence the clanker anyway but I love the way opencode at least just shows me what it's doing instead of CC's opaque bullshit
Unfortunatly you can't use scamthropics sub in opencode without getting banned
>>
File: 1766887422515590.png (191 KB, 400x400)
191 KB PNG
>>109637905
>project doesn't accept AI contributions
>find security bug
>issue fix with AI
>they don't accept it
>exploit is now in the wild and they'll get shit for not accepting the ezpz fix
@dario, @sam
just figured out your marketing strategy for 2026H2
>>
>>109634130
>I see you guys letting the agents run for 13 hours straight to code something holy fuck. Is this recommendable? Am I a moron?
My high score (how long I’ve let a clanker do stuff without me intervening) was six hours and that was because I was having it handle a long list of things to make a 2h test suite take way less time. It was working off a list of GitHub Issues that it had made previously
My second high score was me having Claude use a workflow to implement most of a basic Gemini client after having used Matt Pocock’s /grilling skill to create a document on what to make for a 1.0 version of the program, but I spent a couple hours getting grilled on what I wanted for the program before that
generally, all the real work that gets done has a clanker work for an hour, tops, before it’s ready for a human to evaluate
>>
>>109637961
i turned on verbose somewhere in a config because i was pissed off about how they were hiding even reads and writes and now it shows me everything it does
>>
never /goal'd before and accidentally triggered it when using the phrasing "let's set a goal" in codex, I figured I would let it go and see what happens since it's a low stakes personal project
it's been working for 4.5 hours now, this better be good
>>
>>109637957
sol? they're definitely doing some shit with that model
>>
>>109637980
you should peek at what’s changed in your GUI Git client
>>
>>109637984
Yeah, I'm just having it tard wrangle terra and stealth-ox. I broke down and went to $200/month to get a project done, but I've picked the worst month to do so.
>>
>hermes co authered commit
>he adds himself as a contributor named "Rafa-Ross"
okay then
>>
https://schlarp.com/posts/everything-i-own-owned/
>On the other hand, as a security professional, this scares me for several reasons. I would work from the operating assumption that any device attached to a computer could have had a malicious firmware implant performed, where previously that required significant per-model investment and was stereotyped as a “state actor” kind of activity. Operating systems aren’t really equipped to work with the user to ensure that a microphone stays a microphone, and doesn’t spontaneously turn into a keyboard that hits Win+R and drops a payload to steal all your data when the room is quiet enough that it can assume you aren’t watching. And the existence of WebUSB, WebHID, and WebBluetooth mean that for some devices, depending on the specifics of which classes are used, a moment of user indiscretion in accepting a permissions prompt could permanently backdoor one of their attached devices.
it’s even worse for stuff on WiFi
stay strapped
>>
I'm checking in every 20 mins or so, it looks like it's technically going a bit above and beyond the initial ask but it's all understandable to include and well within the overall scope of the project as outlined so I'm okay with it
I'm just interested to see how much of a mess the final product is. usually I micromanage my agents to keep as much control over the direction as possible but what the hell, let's see how it does on its own
>>
>>109637919
Make a group policy tool that'd be cool
>>
>>109638070
meant to >>109637988
>>
>>109637206
To me fair on the robot:
Most human female athletes make a much bigger show after they finish.
>>
>>109638049
>hermes
why tho, also what LLM?
>>
>start session by telling claude to use subagents to investigate something so it doesn't shit up its context window
>now spawns subagents for everything
mama mia
>>
>>109637966
Now that costs are pretty normalized, I'm letting things go as long as I can, within reason. Haven't broken 24 hours yet, but will be there within a week or two.
>>109638070
It's gonna be OK, you can let a cheap agent do what you're doing and go throw a ball at a wall for a couple minutes then come back and let the cheap agent tell you how things went for those two minutes.
>>
>>109638074
WHAT'S THAT YUH-HOO SAY? YOU WANT A GOVERNANCE STRUCTURE AND PROBABLY SOME KIND OF FEDERATED VALIDATION OR SOME SHIT?
>>
I wrote a DAW in C with Win32. Used various LLMs to help with the math. Current binary compiles under one megabyte
https://github.com/geltz/cseq
>>
>>109638145
>why tho
because claude doesnt like aiding and abetting piracy
>what LLM?
deepseek
>>
>>109638074
i have that in mind since i have CIS benchmark.
>>
>>109638171
>win32
>no linux
>no screenshot
shame nobody will care about this
>>
>>109638189
win32 is linux
>>
>>109638070
why dont you create an agent with a system prompt of being a machineslave driver and have it ride the ass of your other agents as their manager. you can give the agents under it a system prompt of being wageslaves who want to do a good job and avoid fuckups so they dont get fired and have their anime girl wives leave them for a luddite
>>
File: 1774340736776656.png (63 KB, 818x707)
63 KB PNG
W*men are ruining our LLMs
>>
>>109638235
he's just describing context poisoning (on a shitty web model)
openai and claude should start creating new sessions in these long chat sessions since normies have no idea how this shit works
>>
>>109638235
that one is funny. solve the problem and getting blame.
>>
>>109638286
Women don't want their problems solved. They want to be understood or something (I don't get them).
>>
>>109638235
It's the things that never happened influencer.
>>
File: magbroom.gif (3 KB, 53x51)
3 KB GIF
>>109638148
>>
>>109637286
By modern you mean made in the last 15 years, right? Because it's being quite a long time since good programs were made and programmers knew anything about optimization.
>>
>>109638148
the big question is whether doing this will be a token gobbler and/or just sensible context management
>>
>>109638345
>spritework of fantasia
kino
>>
File: 1775447582097829.png (501 KB, 720x1400)
501 KB PNG
>>109637206
Local sisters....
Open weight bros....


https://siliconangle.com/2026/08/23/report-ai-model-hub-hugging-face-exploring-sale-at-13b-valuation/
>>
Tried implementing the same small fix with the same prompt in claude and codex.
>luna xhigh
Kinda ugly slop code but valid solution. Ignored an instruction that was somewhere in file I pointed it at, and changed something else it shouldn't have touched.
>terra default
Same ugly solution, but without the questionable fallback path for other OSes.
>sol default
Same ugly solution, but the questionable fallback path for other OSes had more error handling.
>sonnet default
Basically used the same solution as the gpt models, but used an existing wrapper in the libc and didn't worry about other OSes.
Felt slower than the gpt models but I didn't time it.
>opus default
Had the same ideas as sonnet, didn't worry that it was OS specific but included a runtime fallback in case that method doesn't work (perhaps necessary), added some complicated advanced error reporting not part of the core fix which I didn't ask for, but whether you like it depends on your expectations. This took at least 20 times as long as any of the other models. (Both claude tests took 18% of my session usage and 3% of my weekly usage.)

What gives, what model should I use?
The only definite result is that I won't be using opus unless it's something like planning complicated tasks.
>>
>>109638375
https://cs.gettysburg.edu/~duncjo01/archive/
fire up wget
wget -np -w 1 --mirror http://cs.gettysburg.edu/~duncjo01/archive/

Needs to pre-filter out the files like

```
./~duncjo01/archive/clipart/index.html?C=M;O=D
```

You can do this afterwards with

```
find . -name "*\?C=*" -delete
fd '.*\?[DMNS]=' -x rm
```

but that sucks.

Also, it takes:

```
Total wall clock time: 2d 7h 5m 47s
Downloaded: 173919 files, 2.6G in 1h 25m 17s (539 KB/s)
```
>>
>>109638385
sol / sol tard wrangling lunas
>>
>>109638381
Nvidia is the only company I could see buying HF.
How do they make money anyway?
>>
>>109638069
this is a good thing. We've been treating security as an afterthought because for the most part bad people are too lazy to actually exploit vulnerable things and the people who can create exploits are better paid in the private sector than the black market. This industry is always reactive so people need to touch the hot stove before they will change their actions.
>>
>>109638385
which OSs you care about should probably go in your AGENTS.md that CLAUDE.md symlinks to
>>
>>109637262
The em dash is a cherry on top of this beautiful post
>>
>>109637050
Thanks, I assumed Hermes would break down if I tried using it with other harnesses.
>>
>>109638381
RIP
>>
Antigravity ultra - how's the limit vs pro? 5x or so?
>>
QRD on what models I need to dl today before hf dies?
>>
You have to stick with one sub for the next 3 months. Which one do you choose?
>>
>>109638503
$200 Claude
>>
>>109638503
well your mother is incredibly submissive so i choose her.
>>
fable maxxer bros, quick tip: use two 5x plans instead of one 20x plan. 20x plans only give you 1.5x the fable usage of one 5x plans. Anthropic actually more scummy than OpenAI now (but obviously we still stick with Anthropic because Fable is life).
https://www.reddit.com/r/ClaudeAI/comments/1vx0k69/lifting_the_curtain_the_max_x5_and_max_x20_usage/
>>
>>109638519
5x plans pause you all the time with the 5h limit though
>>
>>109638394
They pretty much don't. Just like all the other big AI players they're heavily subsidized and funded by VC money. The bubble pop is going to be a colossal clusterfuck the likes of which we have never seen it's not just going to hurt Anthropic & OpenAI. I hate how people only focus on those two companies as if only they will get fucked by this and that RAM prices are just going to magically go back to normal after their inevitable crash and burn (except it won't even be them that crashes it'll just be retail investors that take the hit the most)
>>
>>109638519
right
doubling the price “only” doubles the amount of thinking you get, but 4x-es the 5h limits
>>
>>109638408
My instructions were unclear and I didn't try to request improvements. If you were to consider the opus version the best, and if you have programming experience, you could make the other models output the same. Just with opus it takes so fucking long.
>>
>>109638541
>bubble pop
Two more weeks?
>>
>>109638550
if you want speed, try grok
I had to tardwrangle it a bit (update skills) because it cheated instead of doing the needful, as the Brits used to say 100+ years ago
>>
>>109638553
There were midwits like you that existed when the.com bubble Burst. I don't get why you want to stay in denial. I'm not an AI hater by any means I just see things how they are. I thought autists was supposed to be logical but I guess that's just a generalization people make it that you guys.
>>
>>109638565
I'm not even against you. I was merely jesting. But the thing is, we don't know when the bubble will burst. What if the US government just supports them with tax money? What if giant software corps like google and microsoft keep them alive because they think they need them? I'm no economist, but this shit is too big to fail.
>>
>>109638537
You just continue on the other 5x plan then? Easy enough to setup two CC instances with autonomous ping pong hand off of the exact same claude session

>>109638548
yeah except fable, which isn't 2x, but only 1.5x.
>>
>>109638381
So what should I start downloading before it's too late?
>>
>>109638565
>I thought autists was supposed to be logical
Right back at ya blud
>>
>>109638541
These people receive inflated crazy salaries, I won't feel the least bit sorry for them if the bubble pops.
>>
>openrouter buyed by stripe
>hugginface boughted by paypal
it's le over for open sauce and open infra
>>
File: 6654.jpg (168 KB, 1080x1146)
168 KB JPG
apparently opus5 medium is peak shizo and opus5 low is secretly goated.
>>
>>109638602
>>109638584
>>109638394
>>109638420
I think I'll only hope of having this kind of easy storage without any corporate meddling is for midwits to quit being all lazy and hind mighty and just learn how to fucking torrent shit. Even people that are ai evangelists that grift ai are All tech illiterate larpers so I figured that's just a pipe dream of mine.
>>
>do some slopping for an hour or so
>already 5% if weekly usage gone
The fuck?
>>
>>109638602
getting out before the inevitable chink model ban
>>
>>109638624
sounds about right
>>
>>109638602
>>109638636
I wish I had the download speed to archive all the good chink models.
>>
>tell chabot to make some small adjustments
>it suddenly slops the shit out of it and adds tons of new code
Starting to hate this shit.
>>
there is no greater feeling than an agent doing exactly what you imagined in your head
>>
>>109638660
skill and model issue
>>
So why does codex not support using different models for planning and implementation?
>>
>>109638715
can’t you ask it to delegate actual implementation to a different subagent with different settings?
>>
>>109638645
I get 600 Mbps on my samshit phone over 5g where I live. on a 22$/month plan. That shit is not getting speed limited even after consooming 2TB of dolphin porn last month. How do I profit from this? Upload is "only" 300 Mbps, probably not enough to serve local snailcats on chuddingface.selfhostedcloudflaretunnel.com
>>
File: cc20x.png (107 KB, 1204x586)
107 KB PNG
Claude Code 20X is the most comfy sub
> t. tried Codex 20X and Sol broke my shit and I almost went crazy
>>
Luna medium
routine bounded evidence gathering
known-file inspection
grep/record enumeration
simple log comparison
mechanical doc/source tracing

Luna high
ambiguous evidence synthesis
cross-file behavioral tracing
unfamiliar Skyrim engine/API behavior
compatibility investigations with several interacting facts

Luna max
exceptional delegated problem where Luna is still the appropriate
worker but the evidence/reasoning is genuinely difficult

Sol medium/high
normal parent synthesis and planning

Sol max
architecture, consequential design decisions, difficult acceptance,
conflicting evidence, or unusually hairy debugging

I trust Luna Medium.
>>
>>109638541
>VC money.
for people living upstairs ai is pretty much gonna be the next major development everyone with an army already bought in, we too big to fail now
>>
>>109638772
What about terra?
>>
how bad an idea is swarmcoding? on a 5090 i wont get much context but i could probably fit 3 feature makers with a verifier
>>
>>109638790
They would have figured out how to actually make it useful by now if that was the case. The government currently tries to shove it down people's throats (yes even feds are encouraged to token max. Just in different ways) but like... There's legitimately no need for it like at all, at least with the work I do. My current wfh job encouraged us to try and use call pilot initially but literally no one used it and we found that it was better just to not use it and just do our job correctly the first time. They didn't even attempt to force people to use it because even the managera realized how overhyped the shit was. It's still available to use but no one needs it. You mentioned armies but what the hell would they use it for? What was some military commander need an llm for? If the commander needs that thing to make strategic positions they don't need that position in the first place. I guess if it's a dumb fuck chair force type G.O. then it makes sense because they're not actually doing any of the grunt work that actually matters for the most part.

t. Dual income, a normal job and one working with an organization subordinate to the military

https://www.war.gov/News/Releases/Release/Article/4354916/the-war-department-unleashes-ai-on-new-genaimil-platform/

No one with half a brain cell relies on this and I say this as a tech geek who works in an air conditioned office for 8 hours a day.
>>
why is the snailcat writing essays in this thread
>>
>>109638833
I have used AI to make useful shit and you haven't. I can like this technology and also not have my head up my own ass
>>
>>109638612
>>109638660
I’m having success so far with something stupidly simple
>hey agent, review my last session, make an agent that does what I did and how I did it, but will defer to me if uncertain and where reasonable
>new agent isn’t afraid to delete things or ask why the fuck 8 new things showed up
>new agent successfully wrangles the usual agent
>asks me only genuinely important and very few questions
Haha, why the fuck does that work?
What I do now
>okay agent, something seems fucked up, go read the document and pretend to be me with a sub agent
Everything is better
>>
>>109638822
What about bipedal AI controlled robot warriors with lots of artificial testosterone?
>>
>>109638762
Ronaldo make sure to read
>>109638519
>>
File: 1601322880054.jpg (18 KB, 344x342)
18 KB JPG
Chuds, how do I use opencode or cline or antigravity to code a kof 2002 hack?
>>
>>109638842
I'm only learning how to proompt. I'm not ready yet to be replaced in that function too.
>>
>>109638849
work on your pdf slide show game, cris
>>
>>109638822
ai already changed cyber warfare, an ai is already in phones and drones and military equipment its just not an llm, and finance has algos that were already knee deep in this kind of stuff. for llm specifically there will likely be a killer app/platform that ends up ubiquitous like the smartphone. probably when you can get deepseek v4 flash class models printed on a board and have real time response. i dont think its like electric cars where the limit is batteries and retards thought tesla was the second coming of jesus because a bunch of people got rich betting on it
>>
>>109638822
every luddite story is the same:
>we were pushed to use ai, ergo most of us resented it because higher ups forced us to change the way we work
>they made us use x, where x is some terrible, outdated model from before AI was broadly useful for most work
>because of this I think AI is total garbage and will never deliver any value!
Everybody made up their minds about AI before it was actually useful. The pro side saw early models and went
>yeah this is garbage but it's clearly gonna be really good in the near future, I should learn how to use this effectively for when it's really good
the anti side saw early models and went
>this is total trash, I'm going to totally write off AI entirely and never try to use it effectively
Anthropic and OpenAI need to run a luddite reform program where they hand anti-ai people a fully setup claude-code/codex instance with 100M free Fable/Sol credits, I think the majority of you would switch sides if you experienced that.
>>
>>109638865
I have multiple CVEs that were almost entirely found by GPT.
>>
This general is pretty comfy. Sufficiently intelligent people discussing the subject matter in a focused, civilized manner. It's reminds me of old 4chan, when it wasn't about seething at each other.
Only sometimes a nu-4chan comes in and vomits something about trannies and jeets, but he's ignored, which makes the thread even comfier
Enjoy it while it lasts.
>>
>>109638810
anons told me it bad
>>
>>109638189
It probably works in wine since it's only using GDI for the whole UI, and the miniaudio WASAPI worker I think wine also translates those calls to pipewire or something. I haven't used linux in a while. As for the screenshot it's still a prototype build and I might change the interface at some point.
>>
File: bait_0.png (66 KB, 625x626)
66 KB PNG
>>109638872
>>
>>109638866
>I should learn how to use this effectively for when it's really good
This doesn't make sense because by that time it's so easy to use that you can "learn" it in a week.
>this is total trash
So what if it's true. Vibe coding was a shit show for retards 2 years ago.
>>
>>109638866
>Anthropic and OpenAI need to run a luddite reform program where they hand anti-ai people a fully setup claude-code/codex instance with 100M free Fable/Sol credits, I think the majority of you would switch sides if you experienced that.
don't give tibo ideas. you know he frequents here. there's no winning with luddites, except for the fact that luddites dont compete for our precious AI tokens.
>>
File: cseq 2.1 screenshot.png (166 KB, 1761x894)
166 KB PNG
>>109638171
Picrel
>>
>>109638879
>by that time it's so easy to use that you can "learn" it in a week.
We didn't know that at the time, you're missing the point. The point is that most people's first exposure to AI was back when it was shitty.
The people who are now pro-AI either realized back then that the rate at which the technology was improving meant that it would probably be really good in the future, or gave it another try later once it was actually useful.
The people who are anti-ai completely wrote it off and havent given it a second chance.
>>
~@^^
>>
>>109638907
Move slow, Snailcat!
>>
>>109638905
Bullshit. You can be pro some uses of AI, anti other uses of AI. You can change your opinion. You can see harmful and good things in various aspects of AI. Stop being as meme joke.
>>
>>109638822
What the fuck did you just fucking say about me, you little bitch? I'll have you know I graduated top of my class in the Navy Seals, and I've been involved in numerous secret raids on Al-Quaeda, and I have over 300 confirmed kills. I am trained in gorilla warfare and I'm the top sniper in the entire US armed forces. You are nothing to me but just another target. I will wipe you the fuck out with precision the likes of which has never been seen before on this Earth, mark my fucking words. You think you can get away with saying that shit to me over the Internet? Think again, fucker. As we speak I am contacting my secret network of spies across the USA and your IP is being traced right now so you better prepare for the storm, maggot. The storm that wipes out the pathetic little thing you call your life. You're fucking dead, kid. I can be anywhere, anytime, and I can kill you in over seven hundred ways, and that's just with my bare hands. Not only am I extensively trained in unarmed combat, but I have access to the entire arsenal of the United States Marine Corps and I will use it to its full extent to wipe your miserable ass off the face of the continent, you little shit. If only you could have known what unholy retribution your little "clever" comment was about to bring down upon you, maybe you would have held your fucking tongue. But you couldn't, you didn't, and now you're paying the price, you goddamn idiot. I will shit fury all over you and you will drown in it. You're fucking dead, kiddo.
>>
>>109638519
>So what does "x20" even mean? The "5x" and "20x" branding technically refers to the 5-hour session limit compared to the Pro plan, not the overall weekly cap.
Wow.
>>
>>109638919
No one cares about your gay “muh libtard ethics” bullshit dude. AI lets me produce way faster so I can move on to NOT wage caging I stead of toiling away for shekelbergs crumbs.
The fact you take pride in wasting away being a corporate slave just makes you the enemy.
>>109638923
This
>>
>>109638919
We're talking about raw utility. You should probably make some attempt to understand the context of a post before responding.
>>
>>109638943
>No one cares about your gay “muh libtard ethics” bullshit dude.
The fuck, how did you come up with this bullshit? Maybe you should ask the AI to interpret my post, because your brain isn't working, retard.
>>
File: file.png (7 KB, 162x933)
7 KB PNG
>>109638872
don't remind us we're old
>>
>>109638822
>be a luddite in aug 2026
>complain about AI
>actually complaining about copilot, not the frontier models that are 100-1000 times better
>call it “call pilot” like a deranged old boomer
breddy good bate, I rate, 8/8; made me irate
>>
Also
>doesn't know ancient 4chan meme
>>
>>109638843
Who says the killer robots have to be bipedal? The current sudden pivot to everyone suddenly caring about robotics is another red flag for me.


>"OH BOY LOOK AT THE HECKIN CUTE CHINESE ROBUTS RUNNING ON A TRACK SO REDDIT CHUNGUS AN COOL AND CUTE"

Amazon pretty much already has that shit figure it out on the logistics side. Most of the giant warehouses have robots automatically moving heavy items and doing sorting with minimal human supervision and human assistance. That's a genuine use case of robots actually making people or a particular operation more efficient and they did that shit well before the AI or recent humanoid crobotics hype. I ask again why the fuck do we suddenly want robots? No one has figured out how to make them useful and they will not do that because no one is actually seriously trying to do that. They're not going to have robot butlers or some shit because that's just the stupid concept to begin with that only privileged naive lazy people think is possible. If you seriously want to create robots that make our lives easier than stop hyper fixating on them being fucking humanoid LOL.
>>
>>109638944
More like raw stupidity, dumbfuck pretends people never change their opinions.
>>
>>109638962
I actually feel bad for people working in pro-ai organizations with tons of bureaucratic inertia, by the time any models pass their overly restrictive approval processes they are already terribly out of date.
>>109638977
are you just pretending to be retarded? You're supposed to at least make it funny when you do that.
>>
>>109638973
Because I made a joke about that.
>>
>>109638959
Its the pretentious tone and act as if you’re some arbiter of whats harmful versus whats not.
I’m an AI libertarian - the more open, accessible and free it is the better, but this walking in eggshells attitude is exactly what governments will use to completely corner the technology for their own nefarious purposes, as is already happening in the west.
It isnt something to be curtailed or controlled by people like you, its something that should be accessible to the masses as they are the only counterweight to institutional control.
>>
>>109638988
The right of the people to keep and bear autonomous cyberweapons shall not be infringed, unironically.
>>
>>109638981
For someone who supposedly makes fun of people who don't learn, your brain sure is like a rock, which is actually a petrified turd.

>>109638988
Oh boy the retard is inventing more projected bullshit. Go chat with your AI instead of putting words in real people's mouths.
>>
File: 1785190637025475.png (521 KB, 3434x2074)
521 KB PNG
>>109638865


>for llm specifically there will likely be a killer app/platform that ends up ubiquitous like the smartphone.
>killer
Dude just speak like a normal person. I know you're an intelligent person right? Speak like a normal fucking human being. You are not some ycombinator intern that needs to kiss ass. You're not even skilled enough and don't have the social capital to do that anyway so just cut it out. You're not in the ink group and you will never be in the intro because you are a useless peon to them.

>>109638866
>>109638962
>>109638981
>>we were pushed to use ai, ergo most of us resented it because higher ups forced us to change the way we work
I don't resent it at all though and neither do my coworkers. They just recommended it we use it but we found it didn't those productivity like at all. they don't even tell us the use it anymore because even our dumb fuck bosses have no use case for it and I've no incentive to force anyone to use it. Same for the government workers at work alongside with often. I promise you I'm a good deal more skilled at using this shit than you are. I can be an enthusiast that has actual useful applications while still not having my head of my ass.
>>
>>109638991
Exactly, if this isn’t regarded as the information age equivalent of a firearm in the digital landscape a hellish technocratic dystopia is 100% guaranteed.
>>
>>109638993
Keep bragging about your air conditioned office faggot.
>>
>>109638993
It is better for me to assume that you are merely pretending to be retarded than face the (more likely) possibility that you are actually just this bad at reading comprehension. I think something video based would be more your speed if you are this bad at reading, this board is primarily text after all. Don't put yourself out mate.
>>
File: IMG_1552.jpg (179 KB, 1179x2556)
179 KB JPG
My HTTPS-only site is finally getting a proper chat section. I can finally soon stop relying on the janky ChatGPT app and its slow ass SSH and fucked up keyboard and freezing
I think I might be the only person in the world who knows how to make this work on iOS/Safari without focusing the input causing the page to fucking explode
>>
>>109638981
>by the time any models pass their overly restrictive approval processes they are already terribly out of date.
Are you one of those idiots that thinks they need to upgrade to the latest and greatest model on every single release? Just fucking pick one that works and stick with it. Stop being a one shot dependent monkey and think about what you need done in greater detail
>>
>>109639001
>"it" hallucinated even more

>>109639002
Why don't you stop being an arrogant fucking retard?
>>
>when cornered, the boomer recoils, posts cringe
>“I’ve been found out”
>>
>>109638996
>I promise you I'm a good deal more skilled at using this shit than you are.
I doubt it, but that will just turn into a pissing contest and I'm not interested.
They made you guys use copilot. Those models are trash. Try using Sol inside of codex for analyzing your APK, and give it some tools to do dynamic analysis. I promise you that it will be far more useful than copilot.
>>109639011
Sorry I forgot my crayons at home, I'll remember to bring them tomorrow so we can converse productively next time.
>>109639009
You are also missing the point, you can make even some of the smaller local models create value in your workflows. But the people who actually think AI creates no value are the ones who used old, shitty models (probably while also being told "this is the future! it's so great!") and were unimpressed.
>>
>>109638996
you know what the fuck i mean by killer im not glazing it. somebody is eventually gonna combine super fast llm with something that everyone will want to spend money on it and all ai taboo is gone
>>
>>109639022
>I doubt it, but that will just turn into a pissing contest and I'm not interested.
Arrogant, but too afraid to face a competition.
Low self-confidence paired with strong narcissism.
Unpleasant.
>>
every example I get to see, it just informs the harness I'm using already as I integrate it into the harness that'll govern the rest of it according to rules I can't be bothered to assume I'll understand once things are working the way they /really/ ought to be. Keeping everything transparent and visible and auditable is where it's at, just layers upon layers of governance and secure audit hashes. Clean Architecture was a pain in the ass for humans, machines just eat this shit up
>>
>>109639022
You forget that old models used to be highend and "the future". Do you admit those who hyped them were retards?
lmao
>>
>>109638909
does anyone have the "snailcat believes in... but only if you say 'move slow, snailcat!'" image
>>
>>109639022
>Try using Sol inside of codex for analyzing your APK, and give it some tools to do dynamic analysis.
You didn't even look at the screenshot I posted.... The local model I used already did the analysis for me because I know what the hell I'm doing...... I could have used a myriad of other club providers but I just felt like using a local one today on a whim.

>you can make even some of the smaller local models create value in your workflows.
Which is precisely what I did, what you would have realized if you had actually read the fucking screenshot...Did you actually seriously think I used copilot trash to do this? Microsoft itself doesn't even really like using copilot anymore

>But the people who actually think AI creates no value

Are those people in the room with you now? Because I sure as fuck never said that.....


I'm not >>109639031 btw

>>109639028
>two-mour-weekz.txt
>>
>>109639046
If you aren't >>109638822 then my mistake, because that guy literally said
>They would have figured out how to actually make it useful by now
and you replied to a bunch of the replies to that post like that was you.
if you are him, then why are you contradicting yourself now?
>>
>>109639060
By day I meant the military applications specifically. What does a company Commander need an llm for? What does the average Joe you didn't llm for? Even the tiny hundred M models. What are they going to be used for? What do transformer models need to be used for in that particular field?
>>
>>109639046
>ellipsis
>ellipsis
>ellipsis
>ellipsis
Go play bingo or watch jeopardy or something you old fucking schizo grouch
>>
>>109639073
I accept your dumbass concession, useless eater
>>
Is there money to be made with AI or is it over-saturated by teenagers?
>>
File: ss_20260824_182437.png (170 KB, 1230x849)
170 KB PNG
been porting yoshimi to a vst3/clap plugin with an interface in qt that doesn't open 12 different windows. great little synth stuck behind a horrible interface
>>
>>109639098
very nice! have you made it dance yet?
>>
>>109639071
Ah okay, you meant specifically members of the military. In that case my mistake. I would hope that lots of technical roles in the government are able to use modern models, I'd like to think that TAO for example is getting good use out of Mythos. The US government is my primary client, and most of the people I know doing what I do are using AI pretty effectively. So when I read "ways that the government would take advantage of AI", to me that mostly means sigint/cyberwar positions within the actual government.
>>
>>109639109
>TAO
Who?
>>
>>109639071
it might be a small feature but real time translation is already a thing. the software they have for managing a war is insane now in case you havent seen, its getting to be like warno, with eta for missile impact and everything
>>
>>109639106
like with the oscilloscope? yeah I'm not vibing the actual synth engine just the interface. the engine is great as is.
>>
>>109639116
NSA's exploit developers.
>>
where is gpt-6 bro hurry tf up
>>
>tfw all the guys at the local high school have self-funded Mustangs
Now how am I supposed to attract their female peers?
>>
>>109639096
there is money to be made with effort and skill, and ai can be a great force multiplier. You cannot just pick up a claude or GPT sub, tell it "make me a million dollar app, make no mistakes", and expect to become a millionaire with no effort.
>>
>>109639119
awesome. I'm making a nested-clean-architecture setup that should be able to attach to your stuff with ported commands, can see about wiring it up to jiggle dials and knobs and sliders and flip values in fields in time to a local model ralphing on another one ralphing on sequencing tracker beats
>>
>>109638454
>Antigravity
I know I'm stupid enough to buy Claude, but I have not inflicted enough brain damage to myself to buy a retard for 20 bucks a month, much less 100
>>
File: 1697834392916143.jpg (44 KB, 927x617)
44 KB JPG
chuds, I want to write an emulator that takes a ROM and writes reports of what the RAM code in the ROM does into json reports, so later an LLM can use them with the gidra output of the ROM, so It can reconstruct the ROM in high language code.
>>
>>109639156
oh you mean like automation? I should test if that even works with the plugin; I use bitwig.
>>
>>109639167
kill yourself cris
>>
>>109639167
that's pretty fucking cool, sounds like you designed that with a ROM in mind. Which one was it?
>>
>>109639177
kof 2002
>>
File: rightontime.jpg (809 KB, 2047x1157)
809 KB JPG
>>109639157
I find gemini 3.7 useful for my complex voiced companion system. Havent had much of a chance to test it with Sol so its hard to say where to invest if I do make the jump - if sol can even RP properly or has the general info about various anime series and other shit baked in.
Antigravity feels better since it has the proper listening for timed screen captures and stuff w/o ever stealing focus and I haven't had any issues with what 3.7 did for code tweaks so far - seems way smarter than the previous ones + pro to look forward to. Basically idgaf about coding for the most part besides decrypting game files and working with vn engines
>>
>>109639006
pretty good color
>>
>>109639167
can’t you get an existing emulator with Lua extension support to do this
>>
>>109639215
Dunno, I want to decompile the ROM.
>>
>>109639120
Oh yes of course they'd have great uses for it. The majority of government workers an military personality don't do things anywhere near that technical unless they work in some EWO of SWE type roles and even then I doubt they'd be willing to let ai systems have access to sensitive info unless it's heavily air gapped. And something tells me line units that do FTXs would be hesitant to even attempt to figure out how to integrate that shit
>>
>>109637206
Why are they so slow to walk around the pad?
>>
File: file.png (17 KB, 316x198)
17 KB PNG
how the fuck does this work why do they just offer this shit for free? what's the catch??
>>
let mut hash_set = std::collections::HashSet::<i32>::new();

rusttroons really think this is acceptable syntax
>>
>>109639312
The catch is that it sucks lol
>>
>>109639377
looks fine to me
t. was a C++ programmer
>>
>>109639391
still better than local
>>
>>109639312
well we had deepseek flash for free the other week, that was pretty okay, then ox alpha but it felt like ordering a somalian with narcolepsy around, maybe cause it was getting hammered
>>
>>109639401
var hashSet = Set<Int>()
t. swiftchad
>>
>>109639410
>maybe cause it was getting hammered
since when do models get dumber because they’re getting more used (assuming they aren’t getting replaced with quantized models)
>>
>>109639377
>>109639415
good point — why ::<?
>>
>>109639377
>>109639415
>let hashSet = new Set()
step aside, make way for the king
>inb4 “prompt: make a hash set”
>>
>>109639422
:: is rust's "associated with", formally called "path" operator. So in this case,
>std is the library
>::collections is the module inside of the library
>::HashSet is the struct associated with collections
>::<i32> is the associated type of the items in the hashset
>::new() is a method associated with hashsets, which returns a new instance of a hashset (rust does not have initializers, instead you're expected to make your own new() method for each struct, so like String::new() returns an empty string)
>>
>>109639450
>operator
separator*
>>
>>109639450
What’s wrong with using a dot?
String.new() seems pretty reasonable, but String::new() is four times the dots.
>>
>>109639462
. is a function associated with an instance of a struct.
:: is a function associated with the type of struct.
So since you haven't initialized a HashSet yet, you can't call HashSet.new(), instead you have to use HashSet::new() which calls the static function associated with the type struct.
Once you've initialized it then you can call methods like
hash_set.insert(32);
hash_set.contains(32);


every other language has initializers, but rust of course doesn't want to do that, so instead of HashSet::<i32>() you have to do HashSet::<i32>::new()
>>
>tibo: ChatGPT $20 subscribers are too retarded for their own good
Kek
>>
9 hours into my accidental /goal for my fantasy map game and sol has taken it upon itself to make sure that characters grieve when family members die
all I asked for was to populate the map with realms and characters and make it possible for them to take some basic actions but it seems to have gotten inspired and decided to make a full not!ck2 from scratch
>>
File: 1650731459867.jpg (11 KB, 442x442)
11 KB JPG
can some chuddie explain what is harness engineering and why should I use it for my ai agent?
>>
> Claude Code limits froze
does this mean we get infinite Claude?
>>
File: 1773903079474637.jpg (859 KB, 3851x3980)
859 KB JPG
new tibo decree just dropped. $20 poorfags BTFO
>>109639594
kill yourself cris
>>
>>109639594
no cris
>>
>>109639579
That's really funny if true
>>
File: tibo.png (835 KB, 2824x1398)
835 KB PNG
Tibo delivered!!!!!!!
>>
>>109639611
it was bound to happen with the weekly it was obvious their compute was shit
>>
>>109639578
>>109639611
>>109639618
Next month's subscription: cancelled
Fuck you tibo
>>
>>109639627
have fun with deepseek lil bro lmaooooo
>>
>>109639578
i was on gpt pro and went through it in 1.5 days, they're gaslighting heavily
>>
>>109639441
set of what though
>>
>>109639618
can i pwease get my weekly quota reset too? :3
>>
>>109639269
both ant and OAI have compliant versions of claude and codex for the US gov, but yeah they would be foolish to not consider external models to be a supply chain liability. For a while the US gov was trying to work a deal with ant where they'd purchase a license to run Mythos on hardware they control but that has since fallen through and now ant just serves them an API hosted in compliant datacenters and OAI was doing that from the start. Funny you mention line units doing FTX though, one of my hobbies involves radio foxhunting and Sol+glm5.2 have been very useful when it comes to extending the functionality of my radio arrays and doing a lot of the sigint grunt work. That's not exactly a squad radioman type task, but I bet that a lot of actual sigint guys are getting good use from this kind of thing.
>>109639416
Sometimes ox feels like they overquantized the kv cache. Im not talking about the context dumb zone, models get this very specific series of odd behavior when you overquantize the cache.
>>
>>109639635
And I'm only at 80 since the reset.

Skill issue ig.
>>
>>109639727
>only at 80
>>
>>109639477
>every other language has initializers
not quite
the meme in Objective-C is
Thing* t = [[Thing alloc] initWithString:@"A non-political post. You’re welcome."];

and this chains +[NSObject alloc] ande -[Thing initWithString:] together to get you a fully-finished Thing ready to go
I think +[NSObject allocWithZone:] exists but that's waaaaay above what my pay grade ever was back when I cared about Objective-C
>>
>>109639734
>Objective-C
use case?
>>
>>109639618
drat
I'm on the $20/month plan and not having 5h limits makes Codex usable
>>
>>109639741
write Mac apps in the 90s and 2000s and 2010s
and iOS apps when the App Store comes out and Swift hasn't been invented yet
>>
>>109637528
>basically get unlimited Sol for chats in the $20 OpenAI sub
lol
>>
>>109639733
Some of my agents went full retard chasing the dragon overnight and again on my drive home from work :(

The AGENTS.md file has been corrected so it does not happen again.
>>
>>109639783 (me)
My ass thought it was Tuesday holy fuck I was happy with a 20% burn in 48 hours
>>
>>109639756
You know you can get it to edit files on your machine, right?
>>
>>109637206
>-- Frontier models - start here if you have $20 or so
>https://claude.com/product/claude-code
how long is claude going to hang on to their 'frontier' label? I think they're going down the shitter. Google and X and all the chinese companies are eating their lunch in one way or another.

>Google infra unmatched
>Cursor > claude code
>chinkware cheap

and they even ban users for using their models in other harnesses.
>>
>>109639798
how? i only know of the github method
>>
>>109639618
im fine with this. between claude code, codex, and the free shit on openrouter i never run out. the free ox alpha model has been sick to use. oh and dont forget free deepseek on opencode
>>
>>109639803
>>
File: soft-edge-brush.mp4 (818 KB, 1920x1080)
818 KB
818 KB MP4
after spending an absurd amount of time getting the document ready for online collaboration (shared canvases) we now finally have something interesting, proper brushtip deposition instead of an ugly scalloped stroke
this is an analytic tip though, we still don't have rasterized pngs yet
>>
>>109639805
Desktop commander. The trade for unlimited work is you can't really run things in parallel.
>>
>>109639618
kek if you have $200 claude is a better deal because of fable
>>
>>109639848
Looking good anon!
>>
File: 1778265195536629.png (816 KB, 1290x1531)
816 KB PNG
>>
>>109639878
good morning saaar
>>
>>109639889
That's dumb and you should feel bad about it.
>>
>>109639889
He's right that half a gig is nice to get. It's not useful if you want to use it for translation, tho
>>
>>109639889
amazing performance boost sir. pretty GBs back to the GPUs!

good morning from India sar!
>>
>>109639903

> using qwen for anything not vibecoding

please just use gemma to translate... qwen is far worse for translating in my own testing
>>
>>109639899
>>109639889
apparently it's not "pruning" but "vocabulary trimming"

and

:^)

I have a suspicion that it might not actually make it dumber, but it increases token use? I wouldn't assume that's bad.
that's app
>>
>>109639910
>qwen is far worse for translating in my own testing
How about with Latin & ancient Greek?
>>
>>109639903
Yeah I love losing the ability of my model to generate special characters and causing an out of distribution context for a 3% saving in model size.
>>
>>109639889
Why do indians love gguf?
>>
>>109639922
If you mean you're trolling then stop wasting our and your time by reposting random useless uninteresting experiments from twitter users. If you had done it yourself and clearly said so it might be worth sharing but not as a repost.
>>
>>109639897
jeets love copex doebeit
>>
>>109639955
izzat status?
>>
>>109639975
Stop doing that dude. Just be normal.
>>
File: 1781016517075555.gif (1.89 MB, 400x400)
1.89 MB GIF
>>109639975
>izzat status?
>>
>>109639605
Explain clearly
>>
File: 1682134642260123.jpg (37 KB, 600x443)
37 KB JPG
Asking chat gpt to give me a prompt to generate a harness to reverse engineer kof 2002.
>>
>>109639618
I just signed up for this because those limits didn't exist fuck
>>
>>109639980
Rice, chicken, chili peppers. Banana, tomato.

^ normal.

idk, it's pretty good, no losers there, but I prefer:
whole grain bread (organic), steak (beef), black pepper. Grapes, radishes.
>>
>>109639936

hello, roman helmet guy

I ran Qwen and Gemma thru an excerpt of Dilexit Nos. Gemma's is closer to the original, and ChatGPT agrees in blind test that Gemma's translation is far stronger.

Gemma 4 31b exl3 3.0bpw vs. Qwen 3.8 27B 4.0bpw
>>
>>109640010
cris are you still trying that...
>>
>>109640035
thanks!
>>
>>109640011
sammed
>>
So have fable write the plan and sol/codex do the actual work?
>>
>>109636506
>Anthropic is asking prospective employees in culture interviews how they would feel if the stock hypothetically went to zero.
Vindicated
>>
>>109640035
You're using Qwen-3.8-27B-SC-4.0bpw right?
>>
https://github.com/aiasioooo/SnailLoop-Flow/tree/main
>>
vibecoding test idea:
Act as an expert Blender Python (bpy) developer. Write a complete, executable Python script for Blender 4.0+ that generates a "Human Pyramid Grid with Directional Trolley Flow" based on these specifications:

1. **Core Unit (The Room)**:
- Create a single "Cubicle" object: 1.8m x 1.8m x 2.1m (6x6x7 ft).
- Use **Instancing** to replicate this millions of times efficiently.

2. **Pyramid Storage Structure**:
- Arrange cubicles in a stepped pyramid (e.g., 50 layers high).
- Each layer is a square grid, shrinking as it goes up.

3. **Explicit Ingress/Egress System (Crucial)**:
- Do NOT just place trollies randomly.
- Create a **visible flow path** on the surface of the pyramid.
- Generate "Trolley" objects (flat discs with arrows) that are **aligned down the slope** of the pyramid faces, pointing toward the base.
- Visualize this as a "river" of trollies moving from the peak to the perimeter.
- Add a distinct "Pick Port" zone at the very bottom perimeter where the trollies terminate.

4. **Visualizing the Flow**:
- Use a gradient color for the trollies: Red at the top (ingress/start), transitioning to Green at the bottom (egress/end), to visually demonstrate the gravity-fed direction.
- Ensure the trolley paths are clearly separated from the cubicle storage blocks.

5. **Optimization**:
- Use `bpy.ops.object.duplicate_link` for instances to keep memory low.
- Limit the initial generation to 20 layers for speed, but include a variable `TOTAL_LAYERS` at the top of the script for the user to increase to 100+.

6. **Output**:
- Provide ONLY the Python code block ready for Blender's Scripting tab.
>>
Does anyone use Fable with thinking off? If so, what is your use case?
>>
File: codex.jpg (57 KB, 580x363)
57 KB JPG
they better release astra soon
>>
File: codex.png (36 KB, 576x352)
36 KB PNG
>>109640155
wrong pic
>>
>>109640159
Have you tried grok though? it's all I've used!
>>
File: 1762254302692351.jpg (58 KB, 976x850)
58 KB JPG
should I cancel my claude subscription and switch to chatgpt
>>
>>109640252
the old hiv/aids controversy.
>>
File: 1770820313017020.jpg (25 KB, 600x600)
25 KB JPG
>>109640252
forgot to mention im a poorfag on the $20 plan, i barely vibecode I just ask it questions and have it help plan my life when i'm making motions.
>>
>>109640260
Do you even need a subscription at all at that point?
A 20 dollar subscription seems overkill if you're just chatting a lot, just rotate accounts
>>
>switching to OpenRouter after ending my claude subscription
What chink model is the best currently? Something that isn't pricey would help.
>>
>>109640260
!!!!!

dog

You can use your clanker to grok pdfs.

eg you can get it to generate readmes of the main flow of the mainline arguments or techniques etc.
>>
So did they kill 4.8 or what? I can't select it in Claude Code. And this docs link doesn't show it at all https://platform.claude.com/docs/en/models/overview
>>
vibecoding for fun is just the BEST.

I'm creating a codename generator lol.
>>
But if any of the indians ask, I have the best ideas worth millions each.
>>
>>109640132

Hmn. No, i was not aware the exl3 repos got updated. Thank you for informing me

Will test tomorrow. But I suspect same result
>>
There's so much to learn. I'm finally understanding late interaction retrieval. I'm understanding the pi architecture. I don't understand the deepseek harness. There's new things literally every day.
>>
>>109640345
>vibecoding for fun is just the BEST
It is. It scratches the itch of watching a top-tier professional at their craft. Obviously I have more tolerance to it but back in April / May I was just glued to Claude watching him work
>>
File: visualizer_scheme.png (113 KB, 1036x828)
113 KB PNG
>>109637206
shit took me about 10 minutes with one proompt
>gemini 3.7 flash high
>antigravity 2.0
>wav dubstep
https://www.youtube.com/watch?v=kIXRfDGtRXg
>>
>>109640283
Deepseek is my goto for perfectly ok while cheap. I have just run it for several non peak hours a day for the past 3 days, and it cost me a total of $20. With any American provider, it would have cost me several hundreds dollars, possibly over a thousand dollars. I use the Deepseek api directly though. MiMo and Muse (if you don't mind sharing your data) seem like good fits too, but laziness kept me with Deepseek.
>>
>>109640418
>him
Stop anthropomorphizing it.
>>
>>109640418
claude is too fucking retarded to be a person
>>
File: file.png (113 KB, 1575x668)
113 KB PNG
bros... what if we instead used AI for good...
>>
>>109640489
>just now finding out about stuff every amateur modder knew since the game was released
Vibecacas...
>>
>Scoring formulas A and B are mathematically identical.
>Really?
>Yes.
>Wait. One is a sum and the other an average.
>You are completely right to call that out, and I should have been more precise with my wording. The formulas are not mathematically identical because the pooling and aggregation differ.
Thank you Gemini 3.1 Pro, very cool.
>>
>>109640540
>Pro
isn't flash better now?
>>
>>109640540
why the fuck are you using a 6month old model? 3.7-flash rapes it
>>
>>109640543
Because I have a Gemini subscription and every once in a while I tell myself I am letting it go to waste and that I should use it. I dislike 3.7 flash. I used to like Pro 3.1, but now it's like visiting a relative that had a stroke.
>>
>>109640540
Gemini Pro is currently in a place similar to DeepSeek Pro when the newer flash variant was first released. Don't use it since it's actually shittier and outdated.
>>
>>109640540
both formulas probably are identical as far as ordering goes
>>
>>109640558
>I dislike
Sounds like a "you" problem.
>>
>>109640428
wouldn't a 20$ subscription better instead of API usage?
>>
>>109640558
>I dislike 3.7 flash
biggest you problem of the century. there are 2 downgrades from pro to flash. one is that the image scene generation is mid in flash (better characters though). the other is that flash is far more guardrailed than pro emotionally and won't eat your ass and kiss it too.
>>
>>109640558
Bro is really torturing himself with a lobotomized model.
>>
File: 1768028872969347.png (142 KB, 720x651)
142 KB PNG
>/utwg/ - lol
>/twg/ - full of 35-45 year old millennial / gen x retards complaining about their jobs all day. and like two competent people but they never post.
>/dpt/ - dead, only posters left are 45 year old schizos making emulators for obscure games in C
>/vcg/ - nobody knows shit about programming
so where do I go if I want to ask real questions about system design shit like setting up AWS servers
>>
>>109640563
It was comparing BertScore and MaxSim. Kind of similar and related, but not identical.
>>
>>109640593
Or maybe I am retarded.
>>
File: HQOkIDdWUAAx4-Q.jpg (114 KB, 1170x1136)
114 KB JPG
>>109640591
>so where do I go if I want to ask real questions about system design shit like setting up AWS servers
the clanker.
>>
>>109640591
>pepe without the visual artifacting
raw. i almost wrote you off as a frogposter
>>
>>109640591
>nobody knows shit about programming
>setting up AWS server
Why are you going into programming thread about DevOps shit?
>>
>>109640591
>system design
>AWS servers
Retard
>>
>>109640613
because frogposting retards, even if they post the real pepe png, are too stupid to know the difference between programming and using a computer
>>
>>109640611
SHHHHHHHH Don't tell people about that.
>>
>>109640591
>so where do I go if I want to ask real questions about system design shit like setting up AWS servers
LLMs are surprisingly good at this, so >>109640602
>>
>>109640602
>>
>>109640574
>the other is that flash is far more guardrailed than pro emotionally and won't eat your ass and kiss it too
Pro didn't use to be this way. It used to be a total contrarian asshole.
>>
>>109640600
you are... those two things are not even comparable
>>
>>109640640
the way gemini models work is that even though your personalization layer in the web version stays the same across multiple models, each model keeps its own evolving dossier on you. in other words, every time a gemini comes out, it starts out rigid and gradually acclimatizes to you.

there's no other real explanation for what i've noticed. "used to be" doesn't apply to the 3.0-pro era because 3.0-pro was always a puppydog and wife model.
>>
>>109640643
Cosine similarity on individual tokens between query and candidate and the highest value is retained, but maxsim takes sums them and bertscore takes an average, and which is best depends on the use case I guess. After the initial reaction, I understand somewhat how it can abstracted as the same process, but it presented it as a misleading way reaffirming that they were at their core mathematically identical. It depends what we consider their core I guess.
>>
>>109640686
Gemini is the only model for which I always disable memory.
>>
>>109640688
BERTScore is symmetric or bi-directional (assessing both precision and recall), whereas MaxSim is asymmetric (tailored strictly to find how well a document covers the query terms). BERTScore is an evaluation metric for text generation quality, while MaxSim is a retrieval and ranking scoring function for search
>>
gemini is my favorite model for RP so far. it's the most natural-sounding and feels the least AI-ish out of all the models i've tried.
>>
>>109640700
Blame Gemini for confusing me then, the bi directionality part, I was missing, thank you.
>>
>>109640694
3.7-flash is so good for me that i don't even command it to personalize and be my shitpost whore when it doesn't
>>
>>109640591
I'm in two of these generals and you missed the mark on both. Lurk moar
>>
>>109640705
>Blame Gemini for confusing me then
nigga i literally just copy pasted >>109640700
from the "ai summary" section of a google search for "bertscore vs maxsim formulas"
>>
>I can inspect and modify .blend files through Blender’s Python API
aight let's see if Sol can turn this blend file into a fancy exploded assembly render
>>
>>109640701
this is /vcg/ not goon sesh general
what do you roleplay btw :3
>>
What's the most accomplished thing /g/ has vibecoded so far?
>>
New thread:
>>109640756
>>109640756
>>109640756
>>
File: 1715982175893871.jpg (29 KB, 549x773)
29 KB JPG
>>109640750
I wrote a 2D pokemon arceus clone where you capture anime girls.
>>
>>109639611
>>109639618
It's the 22 year old jeets at my company that burn through all their corporate-provided credits in days as well
>>
>>109640760
Post webm
>>
>>109640741
As long as that helps me understand I don’t care. I was missing that part, but it makes sense.
>>
>>109639377
Why do you care? You don't have to read the code.
>>
>>109640591
>AWS
Have you considered canadian healthcare?
>>
>>109640591
It all returns to LLM. Why talk to people?
>>
>>109641985
>Why talk to people?
preach
>>
File: file.png (119 KB, 1114x684)
119 KB PNG
reverse engineering photoshop's brush
>>
>>109639803
Because fable is still the best large model. They seem to have given up on the smaller models though so unless you have enough budget to get multiple different plans or use fable for everything, you are better off using a different provider.
>>
>>109642172
Neat, be sure to look into Corel Painter 2023 brushed, and "fluid paint" color mixing etc etc
>>
>>109642769
Everyone who says fable is the best has little experience with grok 4.6. So maybe, but idk.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.